An improved modal approach for model reduction based on input-output relation

Paper in proceedings, 2012

An approximated model of a large scale dynamical system captures its main features by a proper selection of its dominant modes. In this paper, a new metric is presented for modal dominancy analysis. This metric is an explicit formulation for evaluating the participation of each mode, independent from the other modes, into the output. Only eigensolutions with a large contribution to this metric, the dominant eigensolutions, will be retained in the approximating lower order model. The modal truncation technique which includes the proposed dominancy analysis not only has a metric for error estimation, but is able to automatically detect and eliminate unobservable and uncontrollable modes. An example from finite element analysis of an aluminum plate illustrates the approach and shows that the proposed method is very efficient in respect to computational time.